I found this issue in a locally patched 2.6.32.x, current kernels have
moved the offending code to an __init function which is skipped by
recordmcount.pl, so the bug is not currently being exercised.
However, I think the patch is still a good idea, to avoid future
problems if _mcount were to ever have its address taken in normal
code.
This is what I originally saw:
Although arch/mips/kernel/ftrace.c is built without -pg, and thus
contains no calls to _mcount, it does use the address of _mcount
in ftrace_make_nop(). This was causing relocations to be emitted
for _mcount which recordmcount.pl erronously took to be _mcount
call sites. The result was that the text of ftrace_make_nop()
would be patched with garbage leading to a system crash.
In non-module code, all _mcount call sites will have R_MIPS_26
relocations, so we restrict $mcount_regex to only match on these.
Signed-off-by: David Daney <ddaney@xxxxxxxxxxxxxxxxxx>
Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
Cc: Li Hong <lihong.hi@xxxxxxxxx>
Cc: Ingo Molnar <mingo@xxxxxxx>
Cc: Matt Fleming <matt@xxxxxxxxxxxxxxxxx>
Cc: Ralf Baechle <ralf@xxxxxxxxxxxxxx>
---
scripts/recordmcount.pl | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/scripts/recordmcount.pl b/scripts/recordmcount.pl
index f3c9c0a..0171060 100755
--- a/scripts/recordmcount.pl
+++ b/scripts/recordmcount.pl
@@ -326,7 +326,7 @@ if ($arch eq "x86_64") {
# 14: R_MIPS_NONE *ABS*
# 18: 00020021 nop
if ($is_module eq "0") {
- $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\s_mcount\$";
+ $mcount_regex = "^\\s*([0-9a-fA-F]+): R_MIPS_26\\s+_mcount\$";
} else {
$mcount_regex = "^\\s*([0-9a-fA-F]+): R_MIPS_HI16\\s+_mcount\$";
}
--
1.6.6.1
